Hyne Tumbarumba Trial: Green Board Data
Jonathan Harrington
Todo
- grab a lot more example data for comparison. i.e. typical green board density/stiffness distributions
- more green ecoustic data from Tumbarumba
- more green density/swv data from other sources. DONE (see HPPViSCAN comparison below)
- why is MOE=6.50 over-represented? DONE, 6.5 used as default value
- check that ecoustic MOE is based on \(\rho v^2\) rather than some empirical correlation. DONE, apparently yes
- classify boards as cant or side. DONE
Background
Trimmer
Green board shape could not be collected. USNR software crashed whenever writing shape data to disk enabled.
Data collected USNR database on TRIMMER2.
SICK Camera
Power was lost briefly around 11:40a (someone tripped over the power cord?).
(see sick.Rmd)
Ecoustic
Ecoustic data collected as csv file copied from Ecoustic2 PC by ML.
Loaded into HYNE database using ecoustic-create.sql and ecoustic2db.py.
Ecoustic Dataset
ch = odbcConnect('HYNE','sa','password12')
E = sqlQuery(ch, "select Timestamp, ScheduledDimensions, Width, Thickness, Length, Weight, Density, MOE from ecoustic")
E$swv = sqrt(E$MOE/E$Density*1e9)
E$f1 = E$swv/2/E$Length
#pairs(E[,3:10])
#pairs(E[,c(7,9,8)])
splom(E[,c('Density','swv','MOE')], lower.panel=function(x,y,...){panel.hexbinplot(x,y)})
What causes the cloud of low density high SWV results? Incorrect frequency measured? Different relationship used to determine MOE (remember here SWV is back calculated from MOE and density assuming \(v^2=E/\rho\))
Board properties through time
xyplot(Thickness ~ Timestamp, E)
xyplot(Width ~ Timestamp, E)
xyplot(Length ~ Timestamp, E)
xyplot(Density ~ Timestamp, E)
xyplot(MOE ~ Timestamp, E)
#summary(lm(MOE ~ ScaledMOE, E)) # MOE = ScaledMOEWhat does Length=Thickness=Width=0 denote? All have Grade=‘Not graded’.
Where does the board dimension data come from? The Trimmer2 scanner? Yes, via PLC (not db).
sum(E$Length==0) # 18 length=width=thickness=0## [1] 18sum(E$Density<=0) # 0## [1] NAsum(E$Weight<=0) # 0## [1] 0sum(E$MOE==0) # 122## [1] 122E$MOE[E$Length==0] # always 0## [1] 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0E$MOE[E$Width==0] # always 0## [1] 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0E[E$MOE==6.50,] # nothing unusual## Timestamp ScheduledDimensions Width Thickness Length Weight
## 45 2014-08-31 10:04:50 100x38x7000 98.1 41.3 2425 5.23
## 47 2014-08-31 10:04:56 75x38x7000 76.8 42.1 2425 3.89
## 119 2014-08-31 10:32:03 100x38x7000 98.7 40.8 4850 9.34
## 120 2014-08-31 10:32:04 100x38x7000 98.6 41.0 4850 9.19
## 152 2014-08-31 10:40:48 100x38x7000 97.8 40.7 3637 4.66
## 181 2014-08-31 10:41:54 100x38x7000 98.5 40.7 4850 6.05
## 265 2014-08-31 10:44:50 100x38x7000 98.5 40.7 4850 6.12
## 305 2014-08-31 10:47:09 100x38x7000 98.3 41.2 4850 10.34
## 307 2014-08-31 10:47:11 100x38x7000 97.6 40.7 4850 8.86
## 320 2014-08-31 10:57:46 100x38x7000 97.1 40.7 4243 17.19
## 325 2014-08-31 10:57:50 100x38x7000 98.1 41.4 4850 11.86
## 365 2014-08-31 10:59:15 100x38x7000 98.7 40.7 4921 6.32
## 390 2014-08-31 11:00:13 100x38x7000 97.7 41.0 4850 8.32
## 405 2014-08-31 11:00:23 100x38x7000 98.3 40.7 4850 8.12
## 531 2014-08-31 11:04:00 100x38x7000 101.1 41.3 4243 12.56
## 547 2014-08-31 11:05:32 100x38x7000 98.3 40.8 4850 6.14
## 579 2014-08-31 11:06:04 100x38x7000 99.3 40.8 4850 7.00
## 622 2014-08-31 11:07:03 100x38x7000 98.5 40.7 4850 7.25
## 695 2014-08-31 11:22:01 75x38x7000 77.6 41.8 4850 4.75
## 716 2014-08-31 11:22:31 100x38x7000 97.3 41.0 3027 8.34
## 734 2014-08-31 11:22:52 100x38x7000 98.3 40.4 4850 7.75
## 909 2014-08-31 11:26:05 100x38x7000 98.5 40.7 4850 8.18
## 921 2014-08-31 11:26:13 100x38x7000 97.5 40.7 4850 7.69
## 927 2014-08-31 11:26:17 100x38x7000 97.4 40.5 4850 5.74
## 977 2014-08-31 11:26:53 100x38x7000 98.4 40.7 4850 9.60
## 985 2014-08-31 11:27:00 100x38x7000 98.2 40.3 4850 6.01
## 993 2014-08-31 11:27:11 100x38x7000 98.0 40.4 4850 6.68
## 1012 2014-08-31 11:27:28 100x38x7000 97.4 40.8 4850 7.18
## 1017 2014-08-31 11:27:34 100x38x7000 97.4 40.4 4850 7.23
## 1019 2014-08-31 11:27:36 100x38x7000 97.6 41.2 4850 6.74
## 1020 2014-08-31 11:27:36 100x38x7000 97.9 40.8 4850 6.95
## 1047 2014-08-31 11:28:00 75x38x7000 77.2 40.8 4850 5.21
## 1070 2014-08-31 11:28:14 100x38x7000 98.5 40.7 4850 6.18
## 1136 2014-08-31 11:29:44 100x38x7000 97.8 42.1 4850 13.01
## 1154 2014-08-31 11:30:16 100x38x7000 97.7 40.6 4850 5.76
## 1158 2014-08-31 11:30:19 100x38x7000 98.5 40.7 4860 7.23
## 1199 2014-08-31 11:30:46 100x38x7000 98.3 40.7 4850 7.01
## 1203 2014-08-31 11:30:49 100x38x7000 98.9 40.7 4850 7.04
## 1218 2014-08-31 11:31:35 75x38x7000 76.1 41.6 4850 5.92
## 1346 2014-08-31 11:35:48 100x38x7000 98.3 40.7 4850 6.04
## 1391 2014-08-31 11:37:44 100x38x7000 98.3 40.7 4850 5.29
## 1439 2014-08-31 11:38:30 100x38x7000 98.3 40.8 4850 5.32
## 1446 2014-08-31 11:38:34 100x38x7000 98.8 40.8 4850 5.29
## 1462 2014-08-31 11:38:58 100x38x7000 98.0 40.8 4850 4.69
## 1475 2014-08-31 11:39:09 100x38x7000 98.1 40.7 4850 5.36
## 1537 2014-08-31 11:41:27 100x38x7000 98.3 40.8 4850 5.33
## 1568 2014-08-31 11:41:48 100x38x7000 98.3 40.8 4850 6.83
## 1609 2014-08-31 11:42:17 75x38x7000 76.3 40.5 2435 11.10
## 1612 2014-08-31 11:42:21 100x38x7000 97.2 41.1 4834 7.36
## 1624 2014-08-31 11:42:34 100x38x7000 98.3 40.6 4850 5.15
## 1626 2014-08-31 11:42:35 100x38x7000 97.9 40.3 4850 7.28
## 1627 2014-08-31 11:42:39 100x38x7000 98.4 40.8 4850 5.66
## 1657 2014-08-31 11:43:28 100x38x7000 98.3 40.7 4850 6.06
## 1708 2014-08-31 11:44:31 100x38x7000 98.4 40.8 4850 5.57
## 1720 2014-08-31 11:44:45 100x38x7000 97.2 41.6 4850 7.75
## 1721 2014-08-31 11:44:46 75x38x7000 77.1 41.5 4850 4.64
## 1771 2014-08-31 11:45:47 100x38x7000 98.4 40.7 4850 5.89
## 1797 2014-08-31 11:46:13 100x38x7000 98.4 40.8 4850 5.17
## 1824 2014-08-31 11:46:53 100x38x7000 98.6 40.8 4850 5.68
## 1872 2014-08-31 11:48:13 100x38x7000 98.5 40.6 4850 5.11
## 1901 2014-08-31 11:48:43 100x38x7000 98.1 40.6 4850 10.74
## 1903 2014-08-31 11:48:44 100x38x7000 98.4 40.8 4850 4.92
## 1985 2014-08-31 11:56:50 100x38x7000 98.4 40.8 4850 6.68
## Density MOE swv f1
## 45 532.14 6.5 3494.972 0.7206128
## 47 496.55 6.5 3618.055 0.7459908
## 119 478.00 6.5 3687.591 0.3801640
## 120 468.88 6.5 3723.281 0.3838434
## 152 321.78 6.5 4494.456 0.6178796
## 181 311.09 6.5 4571.026 0.4712398
## 265 315.00 6.5 4542.568 0.4683059
## 305 526.33 6.5 3514.209 0.3622896
## 307 459.88 6.5 3759.537 0.3875812
## 320 1025.09 6.5 2518.116 0.2967376
## 325 602.00 6.5 3285.931 0.3387558
## 365 319.65 6.5 4509.406 0.4581798
## 390 428.23 6.5 3895.993 0.4016487
## 405 418.54 6.5 3940.834 0.4062716
## 531 708.90 6.5 3028.058 0.3568298
## 547 315.85 6.5 4536.451 0.4676754
## 579 356.02 6.5 4272.868 0.4405019
## 622 373.03 6.5 4174.311 0.4303413
## 695 301.71 6.5 4641.537 0.4785090
## 716 690.74 6.5 3067.605 0.5067071
## 734 402.21 6.5 4020.039 0.4144370
## 909 420.98 6.5 3929.397 0.4050925
## 921 399.28 6.5 4034.762 0.4159548
## 927 300.05 6.5 4654.359 0.4798308
## 977 494.39 6.5 3625.950 0.3738093
## 985 313.42 6.5 4554.003 0.4694849
## 993 348.01 6.5 4321.762 0.4455424
## 1012 372.69 6.5 4176.215 0.4305376
## 1017 378.87 6.5 4142.014 0.4270118
## 1019 345.63 6.5 4336.616 0.4470738
## 1020 358.78 6.5 4256.401 0.4388043
## 1047 341.18 6.5 4364.806 0.4499799
## 1070 317.91 6.5 4521.730 0.4661577
## 1136 651.67 6.5 3158.223 0.3255900
## 1154 299.61 6.5 4657.775 0.4801830
## 1158 371.10 6.5 4185.152 0.4305712
## 1199 361.46 6.5 4240.593 0.4371745
## 1203 360.42 6.5 4246.706 0.4378048
## 1218 385.46 6.5 4106.455 0.4233458
## 1346 311.25 6.5 4569.851 0.4711186
## 1391 272.56 6.5 4883.437 0.5034471
## 1439 273.60 6.5 4874.147 0.5024894
## 1446 270.41 6.5 4902.813 0.5054446
## 1462 241.64 6.5 5186.475 0.5346881
## 1475 276.84 6.5 4845.541 0.4995403
## 1537 274.08 6.5 4869.877 0.5020492
## 1568 351.04 6.5 4303.070 0.4436154
## 1609 1475.49 6.5 2098.885 0.4309824
## 1612 380.97 6.5 4130.582 0.4272427
## 1624 266.11 6.5 4942.266 0.5095119
## 1626 380.35 6.5 4133.948 0.4261802
## 1627 290.48 6.5 4730.408 0.4876709
## 1657 312.19 6.5 4562.965 0.4704088
## 1708 286.05 6.5 4766.896 0.4914326
## 1720 395.31 6.5 4054.971 0.4180383
## 1721 299.23 6.5 4660.732 0.4804878
## 1771 303.42 6.5 4628.439 0.4771587
## 1797 265.44 6.5 4948.499 0.5101545
## 1824 291.34 6.5 4723.421 0.4869506
## 1872 263.63 6.5 4965.457 0.5119028
## 1901 555.77 6.5 3419.866 0.3525635
## 1903 252.77 6.5 5071.004 0.5227839
## 1985 343.04 6.5 4352.956 0.4487584Board property distributions
histogram(~Thickness | ScheduledDimensions, E, nint=1000, type='count')
histogram(~Width | ScheduledDimensions, E, nint=1000, type='count')
histogram(~Length | ScheduledDimensions, E, nint=1000, type='count')
histogram(~Weight | ScheduledDimensions, E, nint=100)
histogram(~Density | ScheduledDimensions, E, nint=100)
histogram(~MOE | ScheduledDimensions, E, nint=10001)
~6% of boards have densities>1100 kg/m^3, these are considered unreasonable (a fully saturated board of basic density 500 kg/m^3 has a green density of 1166 kg/m^3).
Why does MOE=6.50 occur way more often than it should?
On the question below there are some cases where 6.5 MoE is assumed. It happens when the confidence in the velocity is low (~50%) but the mill wanted a conservative approach (don’t throw potentially good boards out) so 6.5 is assigned. The real MoE could be above or below that. One reason the confidence could be low on trimmer line 2 is that there can be some bouncing of the test boards away from the hammer when the next large board hits the fence after the trimmer. There is no record in the file of which boards had low confidence.
HPP ViSCAN comparison
TODO: compare current data with ViSCAN data from HPP
Green resonance of 1156 ~6m radiata pine boards measured using Microtec ViSCAN tool. Density from ? Matched to dry bending stiffness.
H = sqlQuery(ch, "select 'viscan' as dataset, greenSWV as swv, greenDEN as Density, dryMOEavg, dryMOElow from HPPviscan")
E$dataset = "ecoustic"
EH = rbind(H[,c("dataset","swv","Density")],E[,c("dataset","swv","Density")])
str(EH)## 'data.frame': 3166 obs. of 3 variables:
## $ dataset: Factor w/ 2 levels "viscan","ecoustic": 1 1 1 1 1 1 1 1 1 1 ...
## $ swv : num 4200 4860 3840 4428 4032 ...
## $ Density: num 597 639 568 582 610 661 584 583 660 660 ...densityplot(~Density, group=dataset, EH, auto.key=TRUE, xlab="Green Density / [kg/m^3]")
ddply(EH, .(dataset), summarize, med=median(Density, na.rm=TRUE))## dataset med
## 1 viscan 964.00
## 2 ecoustic 763.88Both distributions show bimodal density (heartwood, sapwood), but the ecoustic density is substantially lower.
densityplot(~swv/1000, group=dataset, EH, auto.key=TRUE, xlab="Green SWV / [km/s]", subset=swv>0)
ddply(EH, .(dataset), summarize, med=median(swv, na.rm=TRUE))## dataset med
## 1 viscan 3312.000
## 2 ecoustic 3032.886Ecoustic distribution noisier and with a slightly lower mode.
EH$moe=EH$swv^2*EH$Density/1e9
densityplot(~moe, group=dataset, EH, auto.key=TRUE, xlab="Green MOE / [GPa]", subset=swv>0)
ddply(EH, .(dataset), summarize, med=median(moe, na.rm=TRUE))## dataset med
## 1 viscan 9.905753
## 2 ecoustic 6.495000Hmmm. Quite different.
Finally, take a look at the green-dry comparison:
plot(dryMOEavg ~ I(swv^2*Density/1e9), H)
abline(c(0,1), col="red", lty=2)
Remember that a 1st order estimate is:
\[\frac{v_g^2}{v_d^2} = \frac{\rho_d}{\rho_g}\]
Total Recovery
Total flitches seen in trial period (based on TRIMMER2_FlitchDetail table): 1959 (10:29:39–11:56:15 T2.DateTime)
Non trial flitches imaged: 5 (SICK_images.id in 28,33,43,64,102).
Total trial flitches: 1954+1=1955 (extra is due to flitchId=1281565 containing two boards!)
Note that SICK_images.id=43 is FlitchID=1281070 and has no corresponding board!! As do a total of 68 in trial flitches.
Total images containing boards in the SICK_images table: 1886
Total number of boards imaged with SICK camera: 1887. One image contains two boards (sid=1187, 20140831_110531667_GOOD_READ, flitchId=1281565)
Flitches not seen by SICK due to power outage: 73 (11:39:16–11:40:26 T2.DateTime, 11:39–11:40:11 SICK_images.Timestamp)
Total readable images in SICK_barcodes table: 1793 (92% of trial flitches)
Total boards in ecoustic: 1790
Byproduct boards (no ecoustic): 101
Trial boards with no labels: 8
Trial boards with unreadable labels:
ch = odbcConnect('HYNE','sa','password12')
L = sqlQuery(ch, "
select logs.*, t.*, (L.h + S.h)/2. as heartVolFrac from
logs
left join (select t.SWILogNumber, power(heartwoodDiameter_mm,2)/POWER(LED,2) as h from logends as e left join yardTrimmed as t on t.SWILogNumber=e.SWILogNumber where logEnd='L') as L
on L.SWILogNumber=logs.SWILogNumber
left join (select t.SWILogNumber, power(heartwoodDiameter_mm,2)/POWER(SED,2) as h from logends as e left join yardTrimmed as t on t.SWILogNumber=e.SWILogNumber where logEnd='S') as S
on S.SWILogNumber=logs.SWILogNumber
left join yardTrimmed as t
on t.SWILogNumber=logs.SWILogNumber
where logs.SWILogNumber is not null and logs.DateAndTime>'2014-08-31 09:00:00'
order by logs.DateAndTime
")ch = odbcConnect('HYNE','sa','password12')
B.all = sqlQuery(ch, '
select
f.flitchId, f.DateTime as [t.trimmer2], f.SWILogNumber,
f.Width as Wf, f.Thickness as Tf, f.Length as Lf,
f.xc, f.yc, f.R,
b.boardId, b.isByproduct,
b.Width as Wb, b.Thickness as Tb, b.Length as Lb,
s.id as sickImageId, s.Timestamp as [t.SICK], s.nboards, s.filename,
s.boardWidth_pxl*0.137 as Ws, s.boardThickness_pxl*0.137 as Ts,
s.boardOrientation_rad as theta_s,
e.id as ecousticId, e.Timestamp as [t.ecoustic], e.Width, f.Thickness, f.Length,
e.Width as We, e.Thickness as Te, e.Length as Le, e.Weight, e.Density, e.MOE,
pb.name as [sawPattern.boardName],
pb.parent
from
TRIMMER2_FlitchDetail as f
full join
TRIMMER2_BoardDetail as b
on
b.flitchId=f.flitchid
left join
SICK_images as s
on
s.flitchId=f.flitchId
full join
ecoustic as e
on
e.flitchId2=f.flitchId /*why flitchId2?*/
left join
sawPatternBoards as pb
on
pb.flitchId=f.flitchId
where inTrial in (0,1)
order by f.DateTime')
B.all$W = B.all$Wf
B.all$T = B.all$Tf
B.all$L = B.all$Lf
#is100x40 = B.all$Wb>95 & B.all$Tb>36
is100x40 = B.all$Wf>95 & B.all$Tf>35 # 1443 boards
ii = !is100x40 & !is.na(B.all$Le) & B.all$isByproduct!=1 # use ecoustic length except when isByproduct=1
B.all$L[ii] = B.all$Le[ii]
ii = !is100x40 & !is.na(B.all$Lb) & B.all$isByproduct==1 # in which case use Lb
B.all$L[ii] = B.all$Lb[ii]
B.all$V = B.all$W*B.all$T*B.all$L/1e9
B.all$swv = sqrt(B.all$MOE/B.all$Density*1e9)
# total volume recovered as boards
sum(B.all$V)/sum(L$volume) # 67%## [1] 0.6721218# number of boards with known LogNumber
B.knownlog = B.all[!is.na(B.all$SWILogNumber),]
nrow(B.knownlog) # 1754 (90%)## [1] 1954# number of boards with known LogNumber and reasonable MOE and Density (remember MOE=6.50 is a default value)
B = B.all[!B.all$SWILogNumber=='? ' &
!is.na(B.all$SWILogNumber) &
!is.na(B.all$MOE) & B.all$MOE>0 & B.all$MOE!=6.50 &
!is.na(B.all$Density) & B.all$Density<1100 &
B.all$swv<4000,]
B$SWILogNumber = factor(substr(B$SWILogNumber,1,3), ordered=TRUE)
nrow(B) # 1435 (73%)## [1] 1435# number of 100x40's
sum(is100x40) # 1443 (74%)## [1] 1443# number of 100x40's with good green ecoustic data
sum(B$Wf>95 & B$Tf>35) # 1105 boards## [1] 1105# various sets of dodgy logs
marco.drop = c(101, 102, 106, 115, 116, 118, 121, 123, 129, 130, 132, 134, 136, 142, 143, 151, 153, 157, 161, 168, 171, 172, 175, 178, 181, 183, 184, 191, 201)
marco.small.maybe.drop = c(110, 117, 159, 176, 179, 196)
sick.off = c(161, 110, 159, 101, 184, 178, 120, 122, 116, 118, 106, 102, 175, 121, 183, 129)
bad.saw.pattern = c()Finished board dimensions are f.Width, f.Thickness and if 100x40 f.Length, otherwise b.Length (unless b missing…)
Board Dimensions
Let’s compare the dimension data in the trimmer (f=FlitchDetail, b=BoardDetail), SICK (s=skimage.regionprops), ecoustic (e)YY
pairs(B[,c('Wf','Wb','We','Ws')])
pairs(B[,c('Tf','Tb','Te','Ts')])
pairs(B[,c('Lf','Lb','Le')])
imismatch = abs(B$Wf-B$We)>10 | abs(B$Tf-B$Te)>10 | abs(B$Lb-B$Le)>200
B[imismatch,c('flitchId','isByproduct','Wf','Wb','Ws','We','Tf','Tb','Ts','Te','Lf','Lb','Le')]## flitchId isByproduct Wf Wb Ws We Tf Tb Ts Te
## 248 1281309 1 97.5 75 85.22223 98.4 41.3 38 38.03878 40.7
## 358 1281419 1 99.9 75 97.13408 97.6 41.3 38 37.40770 41.4
## 365 1281426 1 97.3 75 98.30507 97.1 28.0 25 27.00420 40.6
## 500 1281561 1 97.0 75 97.03825 76.6 42.7 38 39.17180 40.6
## 516 1281577 1 97.3 75 99.20158 79.6 40.7 38 38.08673 40.4
## 675 1281736 1 97.3 75 98.52973 97.0 42.4 38 43.27950 41.1
## 692 1281753 1 98.0 75 94.38307 97.8 41.6 38 38.11322 41.0
## 736 1281797 1 97.2 75 100.23004 97.2 41.2 38 44.60452 41.2
## 823 1281884 1 99.5 75 96.10237 76.7 41.3 38 40.43689 41.3
## 867 1281928 1 98.1 75 94.55882 97.2 41.8 38 41.80085 40.7
## 1074 1282135 1 98.0 75 94.37286 97.6 41.3 38 37.84145 40.5
## 1112 1282173 1 98.0 75 94.16214 98.4 42.0 38 36.56078 40.9
## 1194 1282255 1 77.3 75 72.17913 97.2 40.9 25 36.52155 40.8
## 1389 1282450 1 77.3 75 72.59385 97.7 43.3 25 40.04053 40.9
## 1438 1282499 1 76.8 75 76.17832 96.7 41.6 25 38.70079 40.7
## 1487 1282548 1 76.5 75 75.66968 97.8 41.6 25 39.33069 40.7
## 1703 1282764 1 97.6 75 76.67522 97.0 42.3 38 37.16935 41.3
## 1782 1282843 1 97.7 75 97.36539 98.4 28.2 25 25.73726 40.7
## 1827 1282888 1 97.7 75 97.16368 77.2 41.5 38 41.25912 40.9
## 1952 1283013 1 98.8 75 99.25655 76.9 42.6 38 39.66019 41.4
## Lf Lb Le
## 248 4925 4200 4942
## 358 4947 4200 4850
## 365 4763 3600 4850
## 500 4906 4800 4850
## 516 4914 4800 4850
## 675 4910 3600 4850
## 692 4924 3600 4850
## 736 4931 4200 4850
## 823 4872 4800 4850
## 867 4914 4200 4850
## 1074 4911 4200 4850
## 1112 4920 3600 4850
## 1194 4928 4800 4850
## 1389 4905 4800 4850
## 1438 4905 3600 4850
## 1487 4935 4800 4850
## 1703 4916 4200 4850
## 1782 4933 4800 4850
## 1827 4891 3600 4850
## 1952 4948 4800 4242The 19 boards (5 of them 100x40s) with significantly different dimensions in the FlitchDetail and Ecoustic tables are all byproducts that normally would have been tippled out and resawn. Instead they were passed through the trim saw and on to the ecoustic which, since it is triggered locally with a PE, tapped them. If the PLC dosn’t know about these boards then the ecoustic acquires the dimensions of the previous board from the PLC. In these cases, the FlitchDetail estimates of dimensions are more likely best, and Density and MOE should be recalculated (but using the BoardDetail length!). In fact, Density and MOE could be recalculated for all isByproduct=1 boards. To do this requires estimating volume from FlitchDetail nominal dimensions (rather than the 3D cloud that the USNR software would have used) and extracting resonant frequency from ecoustic MOE.
AAAARGGGH. Some 100x40’s that the optimiser thought should be re-edged were also trimmed (probably because the do-not-trim logic was based on optimised rather than actual board dimensions) e.g. FlitchId 1282173.
hist(B$L[is100x40],100)
Is log length distinct enough to help identify boards that no good barcode read was extracted from?
bwplot(as.factor(SWILogNumber) ~ Le, B, subset=is100x40)
Board Orientation in SICK Images
ii = !is.na(B$theta_s)
t = B$theta_s[ii]
t[t<0] = pi + t[t<0]
histogram(t/pi*180., breaks=seq(60,120))
summary(t*180/pi)## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 72.06 90.40 91.31 91.41 92.23 112.30Assign SWILogNumber to unidentified boards —————————————–’
Based on:
- sequence
- untrimmed (FlitchDetail) length
- time (but apart from limited cases where whole chain flushed no more info than sequence and a lot harder to work with)
- Saw pattern (see below) but this is pretty tricky
Not much than can be done about the boards 73 boards that passed through while the SICK camera was off (from logs: 161, 110, 159, 101, 184, 178, 120, 122, 116, 118, 106, 102, 175, 121, 183, 129)
plot(1:nrow(B.all), B.all$Lf, cex=0.1)#, xlim=c(0,30))
text(c(1:nrow(B.all)), B.all$Lf+40, B.all$SWILogNumber, cex=0.7)
# Do this using plot.ly (so zoom-pan-identify works easily):
py <- plotly()
trace0 <- list(
#x = B.all$t.trimmer2,
x = 1:nrow(B.all),
y = B.all$Lf,
mode = "markers+text",
text = B.all$SWILogNumber,
textposition = "top",
type = "scatter"
)
layout <- list(
#xaxis = list(range = c(1, 15)),
xaxis = list(range = c(1480, 1580)),
#yaxis = list(range = c(4600, 5000))
yaxis = list(range = c(4850, 4980))
)
response <- py$plotly(trace0, kwargs=list(layout=layout, filename="hyne-untrimmed-board-length", fileopt="overwrite"))
response$url## [1] "https://plot.ly/~letme1n/1"Board Density, MOE and SWV
splom(B[,c('Density','MOE','swv')], lower.panel=function(x,y,...){panel.hexbinplot(x,y)})
Hmmm. What green board data have we got to compare this with?
L = merge(L, ddply(B, .(SWILogNumber), summarize,
nboards=length(SWILogNumber),
boardVolume=sum(V),
boardArea=sum(W*T),
avgBoardMOE = sum(MOE*W*T)/boardArea,
avgBoardDEN = sum(Density*W*T)/boardArea,
avgBoardSWV = sum(swv*W*T)/boardArea),
by="SWILogNumber")
L = merge(L,
ddply(B, .(SWILogNumber), .fun=function (all.boards) {
c=subset(all.boards,substr(all.boards$parent,1,4)=='cant')
if (nrow(c)<1) {
swv = NA
} else {
swv = sum(c$swv*c$W*c$T)/sum(c$W*c$T)
}
return (data.frame(SWILogNumber=all.boards$SWILogNumber[1], nCantBoards=nrow(c), avgCantBoardSWV=swv))
}),
by="SWILogNumber")
L$recFrac = L$boardVolume/L$volume
L$den = L$weight/L$volume
L$MOE = L$SWV^2*L$den / 1e9
L$log.swv = L$SWV # use hitman velocity rather than calibre as there are fewer NA'sxyplot(volume ~ nboards, L)
suspect.lognumbers= c(161, 110, 159, 101, 184, 178, 120, 122, 116, 118, 106, 102, 175, 121, 183, 129) # logs likely to have boards the SIC camera missed during outage
clr = rep('black',nrow(L))
clr[L$SWILogNumber %in% suspect.lognumbers] = 'red'
plot(boardVolume/volume*100 ~ volume, L, ylab='% Recovered Boards with Readable Labels', col=clr)
text(L$volume, L$boardVolume/L$volume*100+1.2, L$SWILogNumber, cex=0.6)
Why is log 137 the best recovered? All boards from 137 are actual reads from the SICK camera rather than maual assignations.
Average Board Density
plot(avgBoardDEN ~ den, L, xlab='Log Density / [kg/m^3]', ylab='Average Green Board Density / [kg/m^3]')
abline(c(0,1),col='grey70')
summary(m <- lm(avgBoardDEN ~ den + heartVolFrac, L)) ##
## Call:
## lm(formula = avgBoardDEN ~ den + heartVolFrac, data = L)
##
## Residuals:
## Min 1Q Median 3Q Max
## -160.374 -37.777 -4.736 40.981 159.981
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -201.4134 163.5470 -1.232 0.221
## den 0.9823 0.1523 6.449 2.46e-09 ***
## heartVolFrac -68.8044 107.1280 -0.642 0.522
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 60.13 on 120 degrees of freedom
## Multiple R-squared: 0.5956, Adjusted R-squared: 0.5889
## F-statistic: 88.38 on 2 and 120 DF, p-value: < 2.2e-16Why is log density so much higher than average board density? Just due to over representation of inner material?
Average Board SWV
Only look at hitman SWV due to issues with calibre SWV (see hyne-log.Rmd).
plot(avgBoardSWV ~ SWV, L, xlab='Log Sonic Velocity / [m/s]', ylab='Average Green Board SWV / [m/s]')
abline(c(0,1),col='grey70')
summary(m <- lm(avgBoardSWV ~ SWV, L))##
## Call:
## lm(formula = avgBoardSWV ~ SWV, data = L)
##
## Residuals:
## Min 1Q Median 3Q Max
## -306.63 -74.07 12.99 75.57 249.26
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -392.26040 148.63533 -2.639 0.00942 **
## SWV 1.04378 0.04508 23.155 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 119.6 on 120 degrees of freedom
## (1 observation deleted due to missingness)
## Multiple R-squared: 0.8171, Adjusted R-squared: 0.8156
## F-statistic: 536.1 on 1 and 120 DF, p-value: < 2.2e-16summary(m <- lm(avgBoardSWV ~ SWV + heartVolFrac, L))##
## Call:
## lm(formula = avgBoardSWV ~ SWV + heartVolFrac, data = L)
##
## Residuals:
## Min 1Q Median 3Q Max
## -304.71 -72.88 13.72 76.04 248.86
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -388.17938 151.84409 -2.556 0.0118 *
## SWV 1.04133 0.04829 21.565 <2e-16 ***
## heartVolFrac 17.54688 120.27446 0.146 0.8843
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 120.1 on 119 degrees of freedom
## (1 observation deleted due to missingness)
## Multiple R-squared: 0.8171, Adjusted R-squared: 0.8141
## F-statistic: 265.9 on 2 and 119 DF, p-value: < 2.2e-16summary(m <- lm(avgBoardSWV ~ SWV + heartVolFrac + den, L)) ##
## Call:
## lm(formula = avgBoardSWV ~ SWV + heartVolFrac + den, data = L)
##
## Residuals:
## Min 1Q Median 3Q Max
## -274.00 -72.95 14.52 70.65 242.59
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 387.82163 345.06958 1.124 0.2633
## SWV 1.04876 0.04736 22.145 <2e-16 ***
## heartVolFrac -435.70662 216.67045 -2.011 0.0466 *
## den -0.74647 0.29957 -2.492 0.0141 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 117.6 on 118 degrees of freedom
## (1 observation deleted due to missingness)
## Multiple R-squared: 0.8263, Adjusted R-squared: 0.8219
## F-statistic: 187.1 on 3 and 118 DF, p-value: < 2.2e-16summary(m <- lm(avgBoardSWV ~ SWV + heartVolFrac + avgBoardDEN, L)) ##
## Call:
## lm(formula = avgBoardSWV ~ SWV + heartVolFrac + avgBoardDEN,
## data = L)
##
## Residuals:
## Min 1Q Median 3Q Max
## -308.86 -71.48 12.03 77.90 216.47
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 350.29967 225.38719 1.554 0.1228
## SWV 0.98172 0.04735 20.732 < 2e-16 ***
## heartVolFrac -359.06919 143.64033 -2.500 0.0138 *
## avgBoardDEN -0.65188 0.15441 -4.222 4.79e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 112.4 on 118 degrees of freedom
## (1 observation deleted due to missingness)
## Multiple R-squared: 0.8411, Adjusted R-squared: 0.8371
## F-statistic: 208.3 on 3 and 118 DF, p-value: < 2.2e-16Average Board MOE
xyplot(avgBoardMOE ~ SWV, L, xlab='Log Sonic Velocity / [m/s]', ylab='Average Green Board MOE / [GPa]')
xyplot(avgBoardMOE ~ MOE, L, xlab='Log MOE / [GPa]', ylab='Average Green Board MOE / [GPa]')
summary(m <- lm(avgBoardMOE ~ MOE, L))##
## Call:
## lm(formula = avgBoardMOE ~ MOE, data = L)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.76947 -0.42905 -0.02078 0.44396 1.55460
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.17282 0.42926 0.403 0.688
## MOE 0.61299 0.04193 14.618 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.6838 on 120 degrees of freedom
## (1 observation deleted due to missingness)
## Multiple R-squared: 0.6404, Adjusted R-squared: 0.6374
## F-statistic: 213.7 on 1 and 120 DF, p-value: < 2.2e-16summary(m <- lm(avgBoardMOE ~ SWV, L))##
## Call:
## lm(formula = avgBoardMOE ~ SWV, data = L)
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.54062 -0.60211 0.02656 0.60749 1.99237
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -3.0188520 1.1257598 -2.682 0.00836 **
## SWV 0.0028588 0.0003414 8.373 1.21e-13 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.906 on 120 degrees of freedom
## (1 observation deleted due to missingness)
## Multiple R-squared: 0.3688, Adjusted R-squared: 0.3635
## F-statistic: 70.11 on 1 and 120 DF, p-value: 1.211e-13summary(m <- lm(avgBoardMOE ~ MOE + den, L))##
## Call:
## lm(formula = avgBoardMOE ~ MOE + den, data = L)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.67513 -0.38362 0.01993 0.39976 1.59725
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -2.0208268 0.8556733 -2.362 0.01982 *
## MOE 0.5820680 0.0420087 13.856 < 2e-16 ***
## den 0.0026832 0.0009144 2.934 0.00401 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.6631 on 119 degrees of freedom
## (1 observation deleted due to missingness)
## Multiple R-squared: 0.6646, Adjusted R-squared: 0.659
## F-statistic: 117.9 on 2 and 119 DF, p-value: < 2.2e-16summary(m <- lm(avgBoardMOE ~ SWV + den, L))##
## Call:
## lm(formula = avgBoardMOE ~ SWV + den, data = L)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.71016 -0.42687 0.02635 0.44095 1.60672
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -1.388e+01 1.388e+00 -10.00 <2e-16 ***
## SWV 3.548e-03 2.642e-04 13.43 <2e-16 ***
## den 9.198e-03 9.357e-04 9.83 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.6758 on 119 degrees of freedom
## (1 observation deleted due to missingness)
## Multiple R-squared: 0.6516, Adjusted R-squared: 0.6458
## F-statistic: 111.3 on 2 and 119 DF, p-value: < 2.2e-16summary(m <- lm(avgBoardMOE ~ MOE + heartVolFrac, L))##
## Call:
## lm(formula = avgBoardMOE ~ MOE + heartVolFrac, data = L)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.6921 -0.4324 -0.0388 0.4639 1.5095
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.69008 0.45383 1.521 0.13102
## MOE 0.60237 0.04088 14.734 < 2e-16 ***
## heartVolFrac -1.80243 0.62579 -2.880 0.00471 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.6639 on 119 degrees of freedom
## (1 observation deleted due to missingness)
## Multiple R-squared: 0.6638, Adjusted R-squared: 0.6582
## F-statistic: 117.5 on 2 and 119 DF, p-value: < 2.2e-16summary(m <- lm(avgBoardMOE ~ SWV + heartVolFrac, L))##
## Call:
## lm(formula = avgBoardMOE ~ SWV + heartVolFrac, data = L)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.73952 -0.46534 -0.09091 0.61570 1.56269
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -4.3726163 0.9322522 -4.690 7.34e-06 ***
## SWV 0.0036727 0.0002965 12.388 < 2e-16 ***
## heartVolFrac -5.8206963 0.7384294 -7.883 1.71e-12 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.7374 on 119 degrees of freedom
## (1 observation deleted due to missingness)
## Multiple R-squared: 0.5853, Adjusted R-squared: 0.5783
## F-statistic: 83.98 on 2 and 119 DF, p-value: < 2.2e-16summary(m <- lm(avgBoardMOE ~ MOE + heartVolFrac + den, L))##
## Call:
## lm(formula = avgBoardMOE ~ MOE + heartVolFrac + den, data = L)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.67622 -0.39895 0.00185 0.45486 1.55828
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -0.882597 1.824861 -0.484 0.630
## MOE 0.589509 0.043395 13.585 <2e-16 ***
## heartVolFrac -0.865620 1.225048 -0.707 0.481
## den 0.001595 0.001792 0.890 0.375
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.6645 on 118 degrees of freedom
## (1 observation deleted due to missingness)
## Multiple R-squared: 0.666, Adjusted R-squared: 0.6576
## F-statistic: 78.45 on 3 and 118 DF, p-value: < 2.2e-16summary(m <- lm(avgBoardMOE ~ SWV + heartVolFrac + den, L))##
## Call:
## lm(formula = avgBoardMOE ~ SWV + heartVolFrac + den, data = L)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.69782 -0.42456 0.01376 0.48863 1.57085
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -1.297e+01 1.989e+00 -6.523 1.80e-09 ***
## SWV 3.590e-03 2.729e-04 13.155 < 2e-16 ***
## heartVolFrac -7.978e-01 1.249e+00 -0.639 0.524
## den 8.272e-03 1.726e-03 4.791 4.86e-06 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.6775 on 118 degrees of freedom
## (1 observation deleted due to missingness)
## Multiple R-squared: 0.6528, Adjusted R-squared: 0.644
## F-statistic: 73.97 on 3 and 118 DF, p-value: < 2.2e-16summary(m <- lm(avgBoardMOE ~ SWV + heartVolFrac + den + avgBoardDEN, L))##
## Call:
## lm(formula = avgBoardMOE ~ SWV + heartVolFrac + den + avgBoardDEN,
## data = L)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1.33707 -0.35998 0.04685 0.41238 1.09554
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -1.352e+01 1.573e+00 -8.596 4.27e-14 ***
## SWV 4.349e-03 2.335e-04 18.629 < 2e-16 ***
## heartVolFrac -1.106e+00 9.873e-01 -1.120 0.265
## den 6.598e-04 1.633e-03 0.404 0.687
## avgBoardDEN 7.468e-03 8.800e-04 8.486 7.67e-14 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.5353 on 117 degrees of freedom
## (1 observation deleted due to missingness)
## Multiple R-squared: 0.7851, Adjusted R-squared: 0.7778
## F-statistic: 106.9 on 4 and 117 DF, p-value: < 2.2e-16Estimates of heartwood volume fraction and log green density add significantly to prediction of average green board stiffness from log SWV.
Saw Patterns
Plot board centroids:
ch = odbcConnect('HYNE','sa','password12')
# get board centroid data - could also do this by pulling xc, yc from TRIMMER2_FlitchDetail
# but doing it this way lets us plot individual barcode information
C = sqlQuery(ch, "select *,
(i1-42)*0.7071*20.0 - (i2-42)*0.2588*20.0 as x,
(i1-42)*0.7071*20.0 + (i2-42)*0.9659*20.0 as y
from SICK_barcodes")# where boardCentroidX_mm is not null")
C$xc = C$boardCentroidX_mm
C$yc = C$boardCentroidY_mm
# get digitization data to overlay on board data
D = sqlQuery(ch, "select flipbookNumber, e.digitizationID, type, x_mm as x, y_mm as y from LogEndDigitizations as d, LogEndDigitizationEdges as e, LogEndDigitizationPoints as p
where p.edgeId=e.id and e.digitizationId=d.id and type='underbark' and logEnd='large'")
D$xrot = cos(1*pi/4)*D$x + sin(1*pi/4)*D$y
D$yrot = -sin(1*pi/4)*D$x + cos(1*pi/4)*D$y
#
xyplot(yc ~ xc | as.factor(SWILogNumber), C,
group=imageId,
aspect='iso', pch=19, type='b',
xlim=c(-1,1)*400, ylim=c(-1,1)*400) +
as.layer(xyplot(y ~ x | as.factor(flipbookNumber), D, group=paste(digitizationID, type),
type='l', aspect='iso', col='grey70')) +
as.layer(xyplot(yrot ~ xrot | as.factor(flipbookNumber), D, group=paste(digitizationID, type),
type='l', aspect='iso', col='grey30'))
SICK camera scale factor
histogram(~pxl2mm, nint=30, C)
summary(C$pxl2mm)## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## 0.1422 0.1553 0.1568 0.1569 0.1582 0.1834 520Does SICK scale factor vary through time?
ch = odbcConnect('HYNE','sa','password12')
F = sqlQuery(ch, "select b.pxl2mm, i.Timestamp from SICK_images as i left join SICK_barcodes as b on b.imageId=i.id where nboards>0")
xyplot(pxl2mm ~ Timestamp, F)
summary(m <- cpt.mean(F$pxl2mm[is.finite(F$pxl2mm)],method='PELT'))## Changepoint type : Change in mean
## Method of analysis : PELT
## Test Statistic : Normal
## Type of penalty : SIC with value, 8.76108
## Maximum no. of cpts : Inf
## Changepoint Locations :plot(m) # no evidence for a shift in mean
Not so much.
The major outlier is image 1187, the image with two boards in one lug. The boards are from two different logs (160, 153) consequently the relative positions are meaningless.
The next worst is 1626. Based on two barcodes. Manual calcs agree, as do calcs based on individual barcodes.
Board orientation w.r.t log csys. Ideally 4 peaks separated by 90 deg. Expect the cant direction (two of the peaks separated by 180 deg) to be more popular.
histogram(~theta*180/pi | as.factor(SWILogNumber), nint=100, C)
Check on board position info
circle = function(r) {
th = seq(0,359)/2/pi
x = r*cos(th)
y = r*sin(th)
return(list(x=x,y=y))
}
xyplot(yc ~ xc | as.factor(SWILogNumber), B, group=paste("R>100",(R>100)), pch=19, panel=function(x,y,...) {
panel.abline(h=0, col="grey70")
panel.abline(v=0, col="grey70")
c = circle(100)
panel.lines(c$x,c$y,lty=2,col='grey70')
panel.xyplot(x,y,...)
}, aspect='iso', auto.key=TRUE)#, subset=SWILogNumber==154)
Some of the edge cases look a little dodgy, but when examined closely seem ok.
Boards with ‘bad’ ecoustic data
Where do the boards in B.all but not in B come from?
xyplot(yc ~ xc | substr(SWILogNumber,1,3), B.all, col='red', cex=1.1) +
as.layer(xyplot(yc ~ xc | SWILogNumber, B, pch=19, col='blue'))
Blue dots represent boards with readable barcodes and ‘good’ ecoustic data (0
As poor a relationship as we saw in the LS15 data. Some logs have a strong pith to bark trend (e.g. 160) but others do not (e.g. 144). Possibly an accurate portrayal of change in MOE with distance from the pith, but also possibly a result of LE position poorly representing whole board position relative to the pith. Might also be influenced by green-ness (will check again when we have dry MoE).
DONE: classify boards as side/cant. see incant.py
xyplot(yc ~ xc | as.factor(SWILogNumber), B, group=parent, pch=19, panel=function(x,y,...) {
panel.abline(h=0, col="grey70")
panel.abline(v=0, col="grey70")
#panel.abline(c(0,tan(cantAngle_rad)))
panel.xyplot(x,y,...)
}, aspect='iso')
xyplot(Density ~ R | as.factor(SWILogNumber), group=parent, B, type="p", subset=parent%in%c('cant1','cant2','cant3'))
Green density patterns look reasonable.
Plot the radial variation of MOE for just the cant boards (colour indicates cant ‘strip’):
xyplot(MOE ~ R | as.factor(SWILogNumber), group=parent, B, type="p",
subset=parent%in%c('cant1','cant2','cant3'), pch=19, cex=1.,
panel=function(...){
panel.abline(h=6.5, col='grey70', lty=3)
panel.xyplot(...)
})
If the ecoustic green density data is unreliable, perhaps better to look at radial variation in green SWV than MOE?
xyplot(swv ~ R | as.factor(SWILogNumber), group=parent, B.all, type="p",
subset=parent%in%c('cant1','cant2','cant3'), pch=19, cex=1.,
panel=function(...){
panel.abline(h=6.5, col='grey70', lty=3)
panel.xyplot(...)
}, ylab="Green SWV / [m/s]", xlab="Board LE Centroid Distance from Pith / [mm]")
As in the total ecoustic SWV data above, numerous very (impossibly?) high SWV values, typically near the pith.
Otherwise, radial MOE variation driven by radial green density variation rather than SWV (maybe?).
Where are the low MOE boards? Remember that MOE=6.5 is a default value.
B$board.class = "<6.5"
B$board.class[B$MOE==6.5] = "=6.5"
B$board.class[B$MOE>6.5] = ">6.5"
B$board.class = factor(B$board.class, levels=c('<6.5','=6.5','>6.5'), ordered=TRUE)
xyplot(yc ~ xc | as.factor(SWILogNumber), B, group=board.class, pch=19, panel=function(x,y,...) {
panel.abline(h=0, col="grey70")
panel.abline(v=0, col="grey70")
#panel.abline(c(0,tan(cantAngle_rad)))
panel.xyplot(x,y,...)
}, aspect='iso', auto.key=TRUE)
Hardly follows the “low MOE boards near the pith, high in the outerwood” paradigm. While some do look like that (e.g. 103, 145, 216), others run completely counter (e.g. 192, 214). Perhaps a consequence of using green swv to predict MOE?
DONE: side board c.f. cant MOE distributions. how many side boards have MOE<6.50?
histogram(~MOE | substr(parent,1,4), B, layout=c(1,2,1), subset=MOE>0&!(B$SWILogNumber%in%marco.drop), panel = function(...) {
panel.histogram(...)
panel.abline(v=6.5, col="red")
})
table(paste("MOE>6.5=",B$MOE>6.5),substr(B$parent,1,4))/nrow(B)##
## cant side
## MOE>6.5= FALSE 0.2125436 0.2146341
## MOE>6.5= TRUE 0.1463415 0.2759582table(paste("MOE>6.5=",B.all$MOE>6.5),substr(B.all$parent,1,4))/nrow(B.all)##
## cant side
## MOE>6.5= FALSE 0.200102093 0.219499745
## MOE>6.5= NA 0.001531394 0.049515059
## MOE>6.5= TRUE 0.139867279 0.21898928050% of side boards have MOE<6.5 c.f. 60% of cant boards.
Improve grade by boxing out pith
In a mill without a cant data, what can be done to improve grade recovery?
DONE: plot probability of getting boards with stiffness<6.5GPa for different cant resawings
Plot the probability a cant board has an MOE of less than 6.5 GPa based on distance from the pith:
C = B[substr(B$parent,1,4)=='cant',]
C = merge(C,L[,c('SWILogNumber','log.swv')],by='SWILogNumber')
C$log.class = 'low.log.swv'
C$log.class[C$log.swv>3306] = 'high.log.swv'
rs = seq(0,300,by=50)
P = data.frame()
for (i in 1:length(rs)) {
P[i,'total'] = sum(C$MOE[C$R>=rs[i]]<6.5)/sum(C$R>=rs[i])
for (c in unique(C$log.class)) {
P[i,c] = sum(C$MOE[C$R>=rs[i] & C$log.class==c]<6.5)/sum(C$R>=rs[i] & C$log.class==c)
}
}
par(mfcol=c(1,1))
matplot(rs,as.matrix(P),type="l",xlab="minimum distance from pith",ylab="fraction boards with MOE<6.5",col=c('black','red','green'),lty=1)
legend('bottomleft',names(P),lty=1,col=c('black','red','green'))
TODO: redraw this plot using distance from the cant outer edges instead of from the pith.
Cant SWV
DONE: Cant SWV from cant boards
DONE: Cant SWV vs log SWV
par(mfcol=c(1,1))
plot(avgCantBoardSWV ~ log.swv, L)
abline(c(0,1), col='grey70', lty=2)
summary(m <- lm(avgCantBoardSWV ~ log.swv, L)) # r^2=0.7##
## Call:
## lm(formula = avgCantBoardSWV ~ log.swv, data = L)
##
## Residuals:
## Min 1Q Median 3Q Max
## -399.93 -105.75 30.33 101.78 392.93
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -325.88347 193.57395 -1.684 0.0951 .
## log.swv 1.02751 0.05858 17.540 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 151.2 on 110 degrees of freedom
## (11 observations deleted due to missingness)
## Multiple R-squared: 0.7366, Adjusted R-squared: 0.7342
## F-statistic: 307.7 on 1 and 110 DF, p-value: < 2.2e-16Relationship between log SWV and average cant board SWV is slightly poorer (r^2=0.7) than same for average swv of all boards (r^2=0.8). Probably due to influence of side boards.
What size are the boards?
B$w = '75'
B$w[B$Width>88] = '100'
B$t = '25'
B$t[B$Thickness>32] = '40'
B$board.size = paste(B$w,B$t,sep='x')
xyplot(yc ~ xc | as.factor(SWILogNumber), B, group=board.size, pch=19, panel=function(x,y,...) {
panel.abline(h=0, col="grey70")
panel.abline(v=0, col="grey70")
#panel.abline(c(0,tan(cantAngle_rad)))
panel.xyplot(x,y,...)
}, aspect='iso', auto.key=TRUE)
Ahh bugger. The cant from log 126 was sawn to 75x40 rather than 100x40. Must’ve been a 225mm cant?
Recommendations
To make merge different board data streams easier:
- include lug id explicitly (is this even possible? if so, who needs to action?)
- leave tipples running and pass byproduct boards back through (though without trimming labelled end) or disable the byproduct option in the optimiser (so PLC knows about all boards)
Assigning log numbers to boards with lost/unreadable labels would be much easier if
- either the chain were flushed between logs
- or logs were trimmed, as squarely as possible, to random but distinctly different lengths (this still dosn’t help with short, waney outer boards though)
If the green and dry board data is to be merged, then to avoid extra effort need to make sure that as many barcodes as possible are read from each dry board SICK image.